Libraries

import time
from itertools import cycle
# dataframes
import pandas as pd
import numpy as np
# Scikit
from sklearn.impute import KNNImputer
from sklearn.preprocessing import MinMaxScaler
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from sklearn.preprocessing import label_binarize
# Metrics
from sklearn.metrics import multilabel_confusion_matrix, confusion_matrix, plot_confusion_matrix
from sklearn.metrics import roc_curve, auc, plot_roc_curve,roc_auc_score
from sklearn.metrics import precision_recall_curve, precision_score, recall_score, f1_score
from sklearn.metrics import plot_precision_recall_curve
from sklearn.metrics import classification_report
# Plot libraries
import matplotlib.pyplot as plt
from venn import venn
import seaborn as sns
# session
from sinfo import sinfo

Functions

def preprocessing(dfTypeR): 
    print("preprocessing")
    dic = {}
    # for loop to go through the rows in the pandas table.
    # It is a preprocessing step.
    for row in dfTypeR.values:
        newRow = []
        # fromp and top are the source and target protein respectively.
        fromp = row[0].split(".")[1]
        top = row[1].split(".")[1]
        # concatenate them into a variable inter.
        inter = fromp+"-"+top
        # same thing for the type of link and their action (if no action the value is "").
        typeR = row[2]+"-"+str(row[3])
        # initialization of keys in <dic>.
        if(inter not in dic):
            dic[inter] = {}
        if(typeR not in dic[inter]):
            dic[inter][typeR] = []
        newRow = []
        # the columns 4 and 5 determine if the relation is directed or not, 
        # in that case, we change the value instead.
        if(row[4] == 'f'):
            newRow.append("Undirected")
            newRow.append(row[6])
            newRow.append(0)
        else:
            if(row[5] =='t'):
                newRow.append("Directed")
                newRow.append(row[6])
                newRow.append(0)
            else:
                newRow.append("Directed")
                newRow.append(row[6])
                newRow.append(1)
        dic[inter][typeR].append(newRow)

    newAr = []
    # We run the dic in a for loop to apply some criterias:
    # 1. if we have directed in both side, we add a new undirected link.
    # 2. the links with actions (except activation or inhibition) we kept.
    # 3. creation of an array to create the new table.
    for k1 in dic:
        vecType = []
        arrP = k1.split("-")
        for k2 in dic[k1]:
            vecType.append(k2)
            countD = 0
            score =0
            for x in dic[k1][k2]:
                if x[0] == "Directed":
                    countD=countD+1
                score = score + x[1]    
            if countD ==2:
                newRows = [["Undirected", int(score/len(dic[k1][k2])),0],
                            ["Directed", int(score/len(dic[k1][k2])),0],
                            ["Directed", int(score/len(dic[k1][k2])),1]]
                dic[k1][k2] = newRows 
            ele = k2
            arrT = k2.split("-")
            if arrT[1] =="" or arrT[1] =="nan" or arrT[0] == "activation" or arrT[0] == "inhibition":
                ele=arrT[0]


            for r in dic[k1][k2]:
                    if(r[2]==1):
                        newAr.append([arrP[1],arrP[0],ele,
                             r[0],r[1]])
                    else:
                        newAr.append([arrP[0],arrP[1],ele,
                             r[0],r[1]])
                        newAr.append([arrP[1],arrP[0],ele,
                             r[0],r[1]])
    print("exporting")
    # transform the array in pandas obj.
    dfTypeR_pre = pd.DataFrame(np.array(newAr), 
                               columns=["Source","Target","Link","Type","Score"])
    print("Done.")
    return(dfTypeR_pre)

def preprocessingV2(dfTypeR): 
    print("preprocessing")
    dic = {}
    # for loop to go through the rows in the pandas table.
    # It is a preprocessing step.
    for row in dfTypeR.values:
        newRow = []
        # fromp and top are the source and target protein respectively.
        fromp = row[0]
        top = row[1]
        # concatenate them into a variable inter.
        inter = fromp+"-"+top
        # same thing for the type of link and their action (if no action the value is "").
        typeR = row[2]+"-"+str(row[3])
        # initialization of keys in <dic>.
        if(inter not in dic):
            dic[inter] = {}
        if(typeR not in dic[inter]):
            dic[inter][typeR] = []
        newRow = []
        # the columns 4 and 5 determine if the relation is directed or not, 
        # in that case, we change the value instead.
        if(row[4] == 'f'):
            newRow.append("Undirected")
            newRow.append(row[6])
            newRow.append(0)
        else:
            if(row[5] =='t'):
                newRow.append("Directed")
                newRow.append(row[6])
                newRow.append(0)
            else:
                newRow.append("Directed")
                newRow.append(row[6])
                newRow.append(1)
        dic[inter][typeR].append(newRow)

    newAr = []
    # We run the dic in a for loop to apply some criterias:
    # 1. if we have directed in both side, we add a new undirected link.
    # 2. the links with actions (except activation or inhibition) we kept.
    # 3. creation of an array to create the new table.
    for k1 in dic:
        vecType = []
        arrP = k1.split("-")
        for k2 in dic[k1]:
            vecType.append(k2)
            countD = 0
            score =0
            for x in dic[k1][k2]:
                if x[0] == "Directed":
                    countD=countD+1
                score = score + x[1]    
            if countD ==2:
                newRows = [["Undirected", int(score/len(dic[k1][k2])),0],
                            ["Directed", int(score/len(dic[k1][k2])),0],
                            ["Directed", int(score/len(dic[k1][k2])),1]]
                dic[k1][k2] = newRows 
            ele = k2
            arrT = k2.split("-")
            if arrT[1] =="" or arrT[1] =="NA" or arrT[0] == "activation" or arrT[0] == "inhibition":
                ele=arrT[0]


            for r in dic[k1][k2]:
                    if(r[2]==1):
                        newAr.append([arrP[1],arrP[0],ele,
                             r[0],r[1]])
                    else:
                        newAr.append([arrP[0],arrP[1],ele,
                             r[0],r[1]])
                        newAr.append([arrP[1],arrP[0],ele,
                             r[0],r[1]])
    print("exporting")
    # transform the array in pandas obj.
    dfTypeR_pre = pd.DataFrame(np.array(newAr), 
                               columns=["Source","Target","Link","Type","Score"])
    print("Done.")
    return(dfTypeR_pre)

##### print("Creating function to split the different link in binary labels")
def transformBinaryLinkTable(tolabelize, df_back,col_pos='Link'):
    df = df_back.copy()
    dic = {}
    for x in df[col_pos].value_counts().index.values:
        if x != tolabelize:
            dic[x] = 0
        else:
            dic[x] = 1
    df[col_pos] = df[col_pos].map(dic)
    df = df.drop_duplicates()
    df['iteraction'] = df.loc[:,'Source'].str.cat(df.loc[:,'Target'],sep="-")
    arr = []
    zero = df.loc[df[col_pos] == 0]
    ones = df.loc[df[col_pos] == 1]
    inter0 = zero['iteraction']
    index0 = inter0.index
    inter1 = ones['iteraction'].to_numpy()
    for idx, interaction in np.ndenumerate(inter0):
        if(interaction in inter1):
            arr.append(index0[idx[0]])

    df = df.drop(arr)
    df = df.loc[:,df.columns != 'iteraction']
    return(df)

Arabidopsis thaliana analysis

org = "athal"
# Reading the preprocess link data.
dfTypeR = pd.read_csv(org+".protein.actions.v11.0.txt.gz",
                      sep="\t",
                      compression="gzip",
                      keep_default_na=True)

# Reading the Seidr file converted in tsv.
dfSf    = pd.read_csv(org+"-network.txt",
                      sep='[;\t]',
                      engine='python',
                      keep_default_na=False,
                      na_values='nan')

# Automatic selection of the colons from Seidr.
selected_col = []
for c in dfSf.columns:
    if "Source" in c or "Target" in c  or "Type" in c or "score" in c  and ("D" not in c and "SDev" not in c):
        selected_col.append(c)
dfSf = dfSf[selected_col]

print("-----------string-action file-----------")
dfTypeR.head()

-----------string-action file-----------

	item_id_a	item_id_b	mode	action	is_directional	a_is_acting	score
0	3702.AT1G01010.1	3702.AT1G01020.1	binding	NaN	f	f	401
1	3702.AT1G01020.1	3702.AT1G01010.1	binding	NaN	f	f	401
2	3702.AT1G01040.2	3702.AT1G01760.1	binding	NaN	f	f	353
3	3702.AT1G01040.2	3702.AT1G07210.1	binding	NaN	f	f	343
4	3702.AT1G01040.2	3702.AT1G07210.1	catalysis	NaN	t	f	343

print("-----------seidr file-----------")
dfSf.head()

-----------seidr file-----------

	Source	Target	Type	ARACNE_score	CLR_score	ELNET_score	GENIE3_score	LLR_score	MI_score	NARROMI_score	PCOR_score	PEARSON_score	TOM_score	PLSNET_score	SPEARMAN_score	TIGRESS_score	irp_score
0	AT1G01225	AT1G01130	Undirected	NaN	2.51912	0.228	0.639124	0.217	0.347712	0.066518	0.001027	0.489676	0.022844	0.002596	0.384108	0.00935	0.267943
1	AT1G01380	AT1G01070	Directed	NaN	1.59224	0.283	0.104246	0.535	0.306454	0.112213	0.003267	0.383988	0.013383	0.008244	0.380991	0.00250	0.315015
2	AT1G01400	AT1G01280	Undirected	NaN	6.35602	0.484	5.355960	0.484	0.333096	0.422627	0.002973	0.929882	0.327482	0.024297	0.730307	0.00190	0.499897
3	AT1G01410	AT1G01280	Directed	NaN	4.67606	0.033	2.155230	0.480	0.306471	NaN	0.002262	0.740514	0.227330	0.015419	0.716679	NaN	0.307832
4	AT1G01410	AT1G01400	Directed	NaN	5.28858	0.223	1.363310	0.474	0.307502	0.180840	0.002149	0.768711	0.207635	0.009831	0.883156	0.00120	0.394951

dfTypeR_pre = preprocessing(dfTypeR)
link_count = dfTypeR_pre['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()
print(dfTypeR_pre)

preprocessing
exporting
Done.

            Source     Target     Link        Type Score
0        AT1G01010  AT1G01020  binding  Undirected   401
1        AT1G01020  AT1G01010  binding  Undirected   401
2        AT1G01020  AT1G01010  binding  Undirected   401
3        AT1G01010  AT1G01020  binding  Undirected   401
4        AT1G01040  AT1G01760  binding  Undirected   353
...            ...        ...      ...         ...   ...
8454543  ATMG00520  ATMG01360  binding  Undirected   607
8454544  ATMG01360  ATMG00730  binding  Undirected   979
8454545  ATMG00730  ATMG01360  binding  Undirected   979
8454546  ATMG01360  ATMG01280  binding  Undirected   704
8454547  ATMG01280  ATMG01360  binding  Undirected   704

[8454548 rows x 5 columns]

to_despreciate = ("ptmod","expression")
dfTypeR_pre = dfTypeR_pre[~dfTypeR_pre['Link'].isin(to_despreciate)]
dfTypeR_pre['Link'] = dfTypeR_pre['Link'].replace(['activation',
                                                   "inhibition",
                                                   "expression-inhibition"],
                                                   'regulation').replace(['binding-activation',
                                                                          'binding-inhibition'],
                                                                          'binding')

# Combine the pandas obj by the columns Source, Target and Type.
result = pd.merge( dfTypeR_pre,dfSf,
                  on=["Source","Target","Type"],
                  how="inner")
# Remove the row duplicates.
result['Type'] = result["Type"].map({"Directed":1,"Undirected":0})
result = result.drop_duplicates()
link_count = result['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

binaryTables = {}
for x in result['Link'].value_counts().index.values:
    print(x)
    binaryTables[x] = transformBinaryLinkTable(x,result)

reaction
binding
catalysis
regulation

interaction= result['Source'].str.cat(result['Target'],sep="-")
ij=pd.DataFrame([interaction, result["Link"]]).T.drop_duplicates()
res = ij.groupby('Link')["Source"].apply(' '.join).reset_index()
dict1 = dict()
for row in res.iterrows():
    dict1[res.Link[row[0]]] = set(res.Source[row[0]].split())
venn(dict1)

<AxesSubplot:>

NAs evaluation

print(result.count())

Source            100939
Target            100939
Link              100939
Type              100939
Score             100939
ARACNE_score        2421
CLR_score         100933
ELNET_score        87459
GENIE3_score      100939
LLR_score         100917
MI_score          100939
NARROMI_score      34472
PCOR_score        100939
PEARSON_score     100939
TOM_score         100939
PLSNET_score      100936
SPEARMAN_score    100939
TIGRESS_score      26333
irp_score         100939
dtype: int64

del result["ARACNE_score"]
del result["NARROMI_score"]
del result["TIGRESS_score"]

X = result.iloc[:,3:]
y = result['Link']
imputer = KNNImputer(n_neighbors=7, weights="uniform")
X_imputed=imputer.fit_transform(X)

X_imputed_abs = pd.DataFrame(X_imputed).abs()
X_imputed_abs.columns = X.columns
print(X_imputed_abs.count())

Type              100939
Score             100939
CLR_score         100939
ELNET_score       100939
GENIE3_score      100939
LLR_score         100939
MI_score          100939
PCOR_score        100939
PEARSON_score     100939
TOM_score         100939
PLSNET_score      100939
SPEARMAN_score    100939
irp_score         100939
dtype: int64

MinMax scaler

scaler = MinMaxScaler()
X_ia_scaled = scaler.fit(X_imputed_abs).transform(X_imputed_abs)
X_ia_scaled = pd.DataFrame(X_ia_scaled)
X_ia_scaled.columns = X_imputed_abs.columns
print(X_ia_scaled)

        Type     Score  CLR_score  ELNET_score  GENIE3_score  LLR_score  \
0        0.0  0.475854   0.225545     0.025045      0.100318   0.902135   
1        0.0  0.598351   0.126708     0.633274      0.035798   0.435943   
2        0.0  0.610130   0.107053     0.692308      0.039033   0.854093   
3        0.0  0.588928   0.123627     0.898032      0.013053   0.870107   
4        0.0  0.610130   0.167283     0.853309      0.009572   0.907473   
...      ...       ...        ...          ...           ...        ...   
100934   1.0  0.243816   0.718279     0.926655      0.470846   0.895018   
100935   0.0  0.778563   0.349276     0.055456      0.187890   0.909253   
100936   0.0  0.537102   0.349276     0.055456      0.187890   0.909253   
100937   0.0  0.537102   0.349276     0.055456      0.187890   0.909253   
100938   1.0  0.243816   0.569066     0.860465      0.425275   0.891459   

        MI_score  PCOR_score  PEARSON_score  TOM_score  PLSNET_score  \
0       0.197339    0.035421       0.690453   0.197234      0.018929   
1       0.103618    0.012421       0.672805   0.123648      0.012784   
2       0.092672    0.022019       0.625273   0.085044      0.023710   
3       0.110479    0.024900       0.629396   0.067731      0.035973   
4       0.124721    0.030389       0.640029   0.121706      0.023755   
...          ...         ...            ...        ...           ...   
100934  0.348724    0.171242       0.889266   0.313348      0.316068   
100935  0.216971    0.125752       0.731314   0.218713      0.197605   
100936  0.216971    0.125752       0.731314   0.218713      0.197605   
100937  0.216971    0.125752       0.731314   0.218713      0.197605   
100938  0.235032    0.194803       0.817496   0.280807      0.397956   

        SPEARMAN_score  irp_score  
0             0.546493   0.186783  
1             0.445894   0.102041  
2             0.425022   0.110152  
3             0.480606   0.150005  
4             0.478759   0.168122  
...                ...        ...  
100934        0.775570   0.588017  
100935        0.670612   0.307715  
100936        0.670612   0.307715  
100937        0.670612   0.307715  
100938        0.632539   0.515519  

[100939 rows x 13 columns]

PCA

X = X_ia_scaled
tn = ['binding','catalysis' ,'reaction', 'regulation']
y = result['Link'].reset_index(drop=True)

Cumulative explained variance

pca = PCA().fit(X)
plt.plot(np.cumsum(pca.explained_variance_ratio_*100))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

Text(0, 0.5, 'cumulative explained variance')

Plot PCAs

pca = PCA(n_components=4, random_state=42)
X_r = pca.fit_transform(X)
plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset (PC1 vs PC4)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 3], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC4: "+str(round(pca.explained_variance_ratio_[3]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Feature importances

i = 1
for com in pca.components_:
    plt.bar(X.columns,com)
    plt.title("PCA " + str(i)) 
    plt.xticks(rotation = 45, ha="right")
    plt.show()
    i+=1

TSNE

tsne= TSNE(n_components=3,init='pca',
                         random_state=42,
                         n_iter = 500 )
X_r = tsne.fit_transform(X)

plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC2")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC2")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Export

pd.concat([y, X_ia_scaled], axis=1).to_csv("results/"+org+".processed_data.tsv",
                               sep='\t',
                               index=False)

Populus tremula analysis

org = "potra"
# Reading the preprocess link data.
dfTypeR = pd.read_csv(org+".protein.actions.v11.0.txt.gz",
                      sep="\t",
                      compression="gzip",
                      keep_default_na=False,
                      na_values='')
# Reading the Seidr file converted in tsv.
dfSf    = pd.read_csv(org+"-network.txt",
                      sep='[;\t]',
                      engine='python',
                      keep_default_na=False,
                      na_values='nan')

# Automatic selection of the colons from Seidr.
selected_col = []
for c in dfSf.columns:
    if "Source" in c or "Target" in c  or "Type" in c or "score" in c  and ("D" not in c and "SDev" not in c):
        selected_col.append(c)
dfSf = dfSf[selected_col]

print("-----------string-action file-----------")
dfTypeR.head()

-----------string-action file-----------

	item_id_a	item_id_b	mode	action	is_directional	a_is_acting	score
0	Potra2n763s36707	Potra2n16c30428	binding	NA	False	False	154
1	Potra2n763s36707	Potra2n5c12706	binding	NA	False	False	154
2	Potra2n1c1035	Potra2n15c28871	binding	NA	False	False	769
3	Potra2n1c1035	Potra2n15c28873	binding	NA	False	False	769
4	Potra2n1c1035	Potra2n1c1053	binding	NA	False	False	400

print("-----------seidr file-----------")
dfSf.head()

-----------seidr file-----------

	Source	Target	Type	ARACNE_score	TOM_score	CLR_score	ELNET_score	GENIE3_score	LLR_score	MI_score	NARROMI_score	PCOR_score	PEARSON_score	PLSNET_score	SPEARMAN_score	TIGRESS_score	irp_score
0	Potra2c131s34686	Potra2c131s34676	Undirected	NaN	0.134824	2.67254	0.182	0.447450	0.316	0.195828	-0.049909	0.003118	0.469940	0.014449	0.396286	0.04980	0.254119
1	Potra2c131s34686	Potra2c131s34677	Undirected	NaN	0.145822	3.39321	0.301	1.201300	0.156	0.278235	0.064569	0.002338	0.525276	0.029884	0.506070	0.06755	0.311845
2	Potra2c131s34686	Potra2c131s34685	Directed	NaN	0.119777	2.03467	0.214	0.754684	0.478	0.298301	-0.074566	0.005393	0.557868	0.059010	0.466023	NaN	0.258337
3	Potra2c131s34687	Potra2c131s34685	Undirected	NaN	0.094158	2.75835	0.478	1.928360	0.496	0.309110	0.082066	0.005953	0.580146	0.072322	0.328183	0.00045	0.354797
4	Potra2c131s34689	Potra2c131s34685	Undirected	NaN	0.085549	2.24622	0.014	0.984184	0.457	0.135046	0.037724	0.004755	0.414607	0.079497	0.141776	NaN	0.249125

dfTypeR_pre = preprocessingV2(dfTypeR)
link_count = dfTypeR_pre['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

preprocessing
exporting
Done.

to_despreciate = ("ptmod","expression")
dfTypeR_pre = dfTypeR_pre[~dfTypeR_pre['Link'].isin(to_despreciate)]
dfTypeR_pre['Link'] = dfTypeR_pre['Link'].replace(['activation',
                                                   "inhibition",
                                                   "expression-inhibition"],
                                                   'regulation').replace(['binding-activation',
                                                                          'binding-inhibition'],
                                                                          'binding')

# Combine the pandas obj by the columns Source, Target and Type.
result = pd.merge( dfTypeR_pre,dfSf,
                  on=["Source","Target","Type"],
                  how="inner")
# Remove the row duplicates.
result['Type'] = result["Type"].map({"Directed":1,"Undirected":0})
result = result.drop_duplicates()
link_count = result['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

binaryTables = {}
for x in result['Link'].value_counts().index.values:
    print(x)
    binaryTables[x] = transformBinaryLinkTable(x,result)

catalysis
binding
reaction
regulation

interaction= result['Source'].str.cat(result['Target'],sep="-")
ij=pd.DataFrame([interaction, result["Link"]]).T.drop_duplicates()
res = ij.groupby('Link')["Source"].apply(' '.join).reset_index()
dict1 = dict()
for row in res.iterrows():
    dict1[res.Link[row[0]]] = set(res.Source[row[0]].split())
venn(dict1)

<AxesSubplot:>

NAs evaluation

print(result.count())

Source            19173
Target            19173
Link              19173
Type              19173
Score             19173
ARACNE_score        515
TOM_score         19173
CLR_score         19167
ELNET_score       19157
GENIE3_score      19173
LLR_score         19091
MI_score          19173
NARROMI_score     11975
PCOR_score        19173
PEARSON_score     19173
PLSNET_score      19173
SPEARMAN_score    19173
TIGRESS_score     12116
irp_score         19173
dtype: int64

del result["ARACNE_score"]

X = result.iloc[:,3:]
y = result['Link']
imputer = KNNImputer(n_neighbors=7, weights="uniform")
X_imputed=imputer.fit_transform(X)

X_imputed_abs = pd.DataFrame(X_imputed).abs()
X_imputed_abs.columns = X.columns
print(X_imputed_abs.count())

Type              19173
Score             19173
TOM_score         19173
CLR_score         19173
ELNET_score       19173
GENIE3_score      19173
LLR_score         19173
MI_score          19173
NARROMI_score     19173
PCOR_score        19173
PEARSON_score     19173
PLSNET_score      19173
SPEARMAN_score    19173
TIGRESS_score     19173
irp_score         19173
dtype: int64

MinMax scaler

scaler = MinMaxScaler()
X_ia_scaled = scaler.fit(X_imputed_abs).transform(X_imputed_abs)
X_ia_scaled = pd.DataFrame(X_ia_scaled)
X_ia_scaled.columns = X_imputed_abs.columns
print(X_ia_scaled)

       Type     Score  TOM_score  CLR_score  ELNET_score  GENIE3_score  \
0       0.0  0.550920   0.370687   0.630766     0.928187      0.339536   
1       1.0  0.887117   0.281385   0.420746     0.786355      0.154908   
2       1.0  0.482209   0.281385   0.420746     0.786355      0.154908   
3       1.0  0.482209   0.281385   0.420746     0.786355      0.154908   
4       1.0  0.482209   0.281385   0.420746     0.786355      0.154908   
...     ...       ...        ...        ...          ...           ...   
19168   1.0  0.611043   0.944016   0.454746     0.962298      0.408790   
19169   1.0  0.776687   0.944016   0.454746     0.962298      0.408790   
19170   1.0  0.776687   0.944016   0.454746     0.962298      0.408790   
19171   0.0  0.587730   0.965266   0.438148     0.834829      0.263687   
19172   0.0  0.587730   0.974376   0.538727     0.861759      0.401984   

       LLR_score  MI_score  NARROMI_score  PCOR_score  PEARSON_score  \
0       0.903052  0.446085       0.229831    0.033544       0.802253   
1       0.064632  0.284923       0.335404    0.054289       0.711291   
2       0.064632  0.284923       0.013917    0.054289       0.711291   
3       0.064632  0.284923       0.013917    0.054289       0.711291   
4       0.064632  0.284923       0.013917    0.054289       0.711291   
...          ...       ...            ...         ...            ...   
19168   0.858169  0.646477       0.489159    0.038419       0.952815   
19169   0.858169  0.646477       0.489159    0.038419       0.952815   
19170   0.858169  0.646477       0.489159    0.038419       0.952815   
19171   0.899461  0.673991       0.155224    0.021159       0.959155   
19172   0.913824  0.767026       0.311852    0.028599       0.918334   

       PLSNET_score  SPEARMAN_score  TIGRESS_score  irp_score  
0          0.022273        0.827896       0.155758   0.299922  
1          0.007832        0.721465       0.014051   0.126421  
2          0.007832        0.721465       0.014051   0.126421  
3          0.007832        0.721465       0.014051   0.126421  
4          0.007832        0.721465       0.014051   0.126421  
...             ...             ...            ...        ...  
19168      0.004459        0.921828       0.527676   0.533342  
19169      0.004459        0.921828       0.527676   0.533342  
19170      0.004459        0.921828       0.527676   0.533342  
19171      0.004096        0.955292       0.068453   0.324774  
19172      0.002441        0.977309       0.043352   0.522864  

[19173 rows x 15 columns]

PCA

X = X_ia_scaled
tn = ['binding','catalysis' ,'reaction', 'regulation']
y = result['Link'].reset_index(drop=True)

Cumulative explained variance

pca = PCA().fit(X)
plt.plot(np.cumsum(pca.explained_variance_ratio_*100))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

Text(0, 0.5, 'cumulative explained variance')

Plot PCAs

pca = PCA(n_components=4, random_state=42)
X_r = pca.fit_transform(X)
plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset (PC1 vs PC4)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 3], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC4: "+str(round(pca.explained_variance_ratio_[3]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Feature importances

i = 1
for com in pca.components_:
    plt.bar(X.columns,com)
    plt.title("PCA " + str(i)) 
    plt.xticks(rotation = 45, ha="right")
    plt.show()
    i+=1

TSNE

tsne= TSNE(n_components=3,init='pca',
                         random_state=42,
                         n_iter = 500 )
X_r = tsne.fit_transform(X)

plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC2")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC2")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Export

pd.concat([y, X_ia_scaled], axis=1).to_csv("results/"+org+".processed_data.tsv",
                               sep='\t',
                               index=False)

Eucaliptus grandis analysis

org = "eugra"
# Reading the preprocess link data.
dfTypeR = pd.read_csv(org+"V2.protein.actions.v11.0.txt",
                      sep="\t",
                      keep_default_na=False,
                      na_values='')
# Reading the Seidr file converted in tsv.
dfSf    = pd.read_csv(org+"-network.txt",
                      sep='[;\t]',
                      engine='python',
                      keep_default_na=False,
                      na_values='nan')

# Automatic selection of the colons from Seidr.
selected_col = []
for c in dfSf.columns:
    if "Source" in c or "Target" in c  or "Type" in c or "score" in c  and ("D" not in c and "SDev" not in c):
        selected_col.append(c)
dfSf = dfSf[selected_col]

print("-----------string-action file-----------")
dfTypeR.head()

-----------string-action file-----------

	item_id_a	item_id_b	mode	action	is_directional	a_is_acting	score
0	Eucgr.H02311	Eucgr.A01152	catalysis	NA	True	False	260
1	Eucgr.H02311	Eucgr.H01473	catalysis	NA	True	True	268
2	Eucgr.H02311	Eucgr.A00913	reaction	NA	True	True	625
3	Eucgr.H02311	Eucgr.B03575	binding	NA	False	False	264
4	Eucgr.H02311	Eucgr.B03575	catalysis	NA	True	False	254

print("-----------seidr file-----------")
dfSf.head()

-----------seidr file-----------

	Source	Target	Type	ARACNE_score	CLR_score	ELNET_score	GENIE3_score	LLR_score	MI_score	NARROMI_score	PCOR_score	PEARSON_score	PLSNET_score	SPEARMAN_score	TIGRESS_score	irp_score
0	Eucgr.A00081	Eucgr.A00054	Undirected	NaN	3.80515	NaN	0.007638	NaN	0.308182	0.314152	-0.000073	-0.890007	0.007300	-0.847375	0.02450	0.266487
1	Eucgr.A00092	Eucgr.A00018	Undirected	NaN	5.53434	NaN	0.014977	NaN	0.235245	0.204778	-0.000020	-0.800503	0.096138	-0.807082	0.02995	0.266311
2	Eucgr.A00189	Eucgr.A00039	Undirected	NaN	4.06924	NaN	0.010511	0.449	0.184084	0.148266	0.000297	0.723548	0.104756	0.551061	0.00670	0.259268
3	Eucgr.A00197	Eucgr.A00096	Undirected	NaN	5.10176	NaN	0.017955	0.001	0.221552	0.220805	-0.000430	-0.817597	0.232075	-0.722833	0.18875	0.320857
4	Eucgr.A00240	Eucgr.A00038	Undirected	NaN	4.78214	NaN	0.044025	0.001	0.247125	NaN	0.000384	0.826510	0.025068	0.838828	0.14025	0.259200

dfTypeR_pre = preprocessingV2(dfTypeR)
link_count = dfTypeR_pre['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

preprocessing
exporting
Done.

to_despreciate = ("ptmod","expression")
dfTypeR_pre = dfTypeR_pre[~dfTypeR_pre['Link'].isin(to_despreciate)]
dfTypeR_pre['Link'] = dfTypeR_pre['Link'].replace(['activation',
                                                   "inhibition",
                                                   "expression-inhibition"],
                                                   'regulation').replace(['binding-activation',
                                                                          'binding-inhibition'],
                                                                          'binding')

# Combine the pandas obj by the columns Source, Target and Type.
result = pd.merge( dfTypeR_pre,dfSf,
                  on=["Source","Target","Type"],
                  how="inner")
# Remove the row duplicates.
result['Type'] = result["Type"].map({"Directed":1,"Undirected":0})
result = result.drop_duplicates()
link_count = result['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

binaryTables = {}
for x in result['Link'].value_counts().index.values:
    print(x)
    binaryTables[x] = transformBinaryLinkTable(x,result)

catalysis
binding
reaction
regulation

interaction= result['Source'].str.cat(result['Target'],sep="-")
ij=pd.DataFrame([interaction, result["Link"]]).T.drop_duplicates()
res = ij.groupby('Link')["Source"].apply(' '.join).reset_index()
dict1 = dict()
for row in res.iterrows():
    dict1[res.Link[row[0]]] = set(res.Source[row[0]].split())
venn(dict1)

<AxesSubplot:>

NAs evaluation

print(result.count())

Source            418
Target            418
Link              418
Type              418
Score             418
ARACNE_score       25
CLR_score         418
ELNET_score       166
GENIE3_score      406
LLR_score         125
MI_score          418
NARROMI_score     288
PCOR_score        418
PEARSON_score     418
PLSNET_score      418
SPEARMAN_score    418
TIGRESS_score     417
irp_score         418
dtype: int64

del result["ARACNE_score"]

X = result.iloc[:,3:]
y = result['Link']
imputer = KNNImputer(n_neighbors=7, weights="uniform")
X_imputed=imputer.fit_transform(X)

X_imputed_abs = pd.DataFrame(X_imputed).abs()
X_imputed_abs.columns = X.columns
print(X_imputed_abs.count())

Type              418
Score             418
CLR_score         418
ELNET_score       418
GENIE3_score      418
LLR_score         418
MI_score          418
NARROMI_score     418
PCOR_score        418
PEARSON_score     418
PLSNET_score      418
SPEARMAN_score    418
TIGRESS_score     418
irp_score         418
dtype: int64

MinMax scaler

scaler = MinMaxScaler()
X_ia_scaled = scaler.fit(X_imputed_abs).transform(X_imputed_abs)
X_ia_scaled = pd.DataFrame(X_ia_scaled)
X_ia_scaled.columns = X_imputed_abs.columns
print(X_ia_scaled)

     Type     Score  CLR_score  ELNET_score  GENIE3_score  LLR_score  \
0     0.0  0.282367   0.447032     0.169592      0.002169   0.232426   
1     0.0  0.045623   0.196300     0.079395      0.048146   0.336168   
2     0.0  0.133169   0.354474     0.019444      0.460900   0.081066   
3     1.0  0.686806   0.372963     0.559546      0.032456   0.141723   
4     1.0  0.675709   0.372963     0.559546      0.032456   0.168651   
..    ...       ...        ...          ...           ...        ...   
413   1.0  0.616523   0.426321     0.661356      0.081493   0.184524   
414   1.0  0.685573   0.447088     0.419930      0.013688   0.157029   
415   1.0  0.030826   0.527551     0.223062      0.000096   0.156746   
416   0.0  0.002466   0.834629     0.003511      0.185017   0.676871   
417   1.0  0.924784   0.758550     0.784499      0.043015   0.121032   

     MI_score  NARROMI_score  PCOR_score  PEARSON_score  PLSNET_score  \
0    0.546079       0.405223    0.032253       0.761440      0.009525   
1    0.167276       0.220093    0.168772       0.499497      0.120955   
2    0.788794       0.614186    0.095319       0.901537      0.105301   
3    0.362053       0.295411    0.032258       0.814806      0.021006   
4    0.362053       0.145091    0.032258       0.814806      0.021006   
..        ...            ...         ...            ...           ...   
413  0.420990       0.120127    0.388626       0.755045      0.177902   
414  0.399616       0.222039    0.095130       0.659602      0.095937   
415  0.271132       0.174762    0.162373       0.538943      0.207802   
416  0.529411       0.449941    0.496397       0.800826      0.314147   
417  0.448876       0.251721    0.252345       0.746388      0.368515   

     SPEARMAN_score  TIGRESS_score  irp_score  
0          0.827150       0.010013   0.217586  
1          0.626677       0.000693   0.064570  
2          0.745857       0.046977   0.471757  
3          0.868982       0.053589   0.236089  
4          0.868982       0.053589   0.236089  
..              ...            ...        ...  
413        0.898877       0.147670   0.366585  
414        0.759273       0.023804   0.120738  
415        0.629045       0.019710   0.119344  
416        0.493292       0.024370   0.369902  
417        0.388319       0.036965   0.363575  

[418 rows x 14 columns]

PCA

X = X_ia_scaled
tn = ['binding','catalysis' ,'reaction', 'regulation']
y = result['Link'].reset_index(drop=True)

Cumulative explained variance

pca = PCA().fit(X)
plt.plot(np.cumsum(pca.explained_variance_ratio_*100))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

Text(0, 0.5, 'cumulative explained variance')

Plot PCAs

pca = PCA(n_components=4, random_state=42)
X_r = pca.fit_transform(X)
plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset (PC1 vs PC4)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 3], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC4: "+str(round(pca.explained_variance_ratio_[3]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Feature importances

i = 1
for com in pca.components_:
    plt.bar(X.columns,com)
    plt.title("PCA " + str(i)) 
    plt.xticks(rotation = 45, ha="right")
    plt.show()
    i+=1

TSNE

tsne= TSNE(n_components=3,init='pca',
                         random_state=42,
                         n_iter = 500 )
X_r = tsne.fit_transform(X)

plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC2")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC2")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Export

pd.concat([y, X_ia_scaled], axis=1).to_csv("results/"+org+".processed_data.tsv",
                               sep='\t',
                               index=False)

Drosophila melanogaster analysis

org = "dmel"
# Reading the preprocess link data.
dfTypeR = pd.read_csv(org+".protein.actions.v11.0.txt",
                      sep="\t",
                      keep_default_na=False,
                      na_values='')
# Reading the Seidr file converted in tsv.
dfSf    = pd.read_csv(org+"-network.txt",
                      sep='[;\t]',
                      engine='python',
                      keep_default_na=False,
                      na_values='nan')

# Automatic selection of the colons from Seidr.
selected_col = []
for c in dfSf.columns:
    if "Source" in c or "Target" in c  or "Type" in c or "score" in c  and ("D" not in c and "SDev" not in c):
        selected_col.append(c)
dfSf = dfSf[selected_col]

print("-----------string-action file-----------")
dfTypeR.head()

-----------string-action file-----------

	item_id_a	item_id_b	mode	action	is_directional	a_is_acting	score
0	FBgn0031089	FBgn0039955	activation	activation	False	False	243
1	FBgn0031089	FBgn0004837	activation	activation	False	False	309
2	FBgn0031089	FBgn0004837	binding	NA	False	False	311
3	FBgn0031092	FBgn0023516	binding	NA	False	False	952
4	FBgn0031092	FBgn0023516	reaction	NA	False	False	952

print("-----------seidr file-----------")
dfSf.head()

-----------seidr file-----------

	Source	Target	Type	ARACNE_score	CLR_score	ELNET_score	GENIE3_score	LLR_score	MI_score	NARROMI_score	PCOR_score	PEARSON_score	TOMSimilarity_score	PLSNET_score	SPEARMAN_score	TIGRESS_score	irp_score
0	FBgn0000014	FBgn0000015	Directed	1.04682	9.88027	0.508	45.6294	0.504	1.046820	0.581011	0.016220	0.860236	0.053560	0.020510	0.860757	1.00000	0.620797
1	FBgn0000036	FBgn0000037	Directed	NaN	7.13729	0.427	36.4574	0.175	0.967644	0.267959	0.002888	0.859128	0.264029	0.003396	0.827299	NaN	0.341291
2	FBgn0000038	FBgn0000036	Undirected	NaN	4.52725	0.004	35.4361	0.396	0.792940	0.200561	0.001564	0.795708	0.241953	0.007585	0.704159	NaN	0.296957
3	FBgn0000038	FBgn0000037	Undirected	NaN	5.93120	0.258	23.0094	0.449	0.939542	0.259599	0.003665	0.852597	0.305338	0.008410	0.742275	NaN	0.348288
4	FBgn0000039	FBgn0000036	Undirected	1.02344	7.33673	0.505	41.9977	0.508	1.023440	NaN	0.003119	0.883952	0.256364	0.004346	0.803889	0.01015	0.449333

dfTypeR_pre = preprocessingV2(dfTypeR)
link_count = dfTypeR_pre['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

preprocessing
exporting
Done.

to_despreciate = ("ptmod","expression")
dfTypeR_pre = dfTypeR_pre[~dfTypeR_pre['Link'].isin(to_despreciate)]
dfTypeR_pre['Link'] = dfTypeR_pre['Link'].replace(['activation',
                                                   "inhibition",
                                                   "expression-inhibition"],
                                                   'regulation').replace(['binding-activation',
                                                                          'binding-inhibition'],
                                                                          'binding')

# Combine the pandas obj by the columns Source, Target and Type.
result = pd.merge( dfTypeR_pre,dfSf,
                  on=["Source","Target","Type"],
                  how="inner")
# Remove the row duplicates.
result['Type'] = result["Type"].map({"Directed":1,"Undirected":0})
result = result.drop_duplicates()
link_count = result['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

binaryTables = {}
for x in result['Link'].value_counts().index.values:
    print(x)
    binaryTables[x] = transformBinaryLinkTable(x,result)

catalysis
binding
reaction
regulation

interaction= result['Source'].str.cat(result['Target'],sep="-")
ij=pd.DataFrame([interaction, result["Link"]]).T.drop_duplicates()
res = ij.groupby('Link')["Source"].apply(' '.join).reset_index()
dict1 = dict()
for row in res.iterrows():
    dict1[res.Link[row[0]]] = set(res.Source[row[0]].split())
venn(dict1)

<AxesSubplot:>

NAs evaluation

print(result.count())

Source                 17416
Target                 17416
Link                   17416
Type                   17416
Score                  17416
ARACNE_score            1094
CLR_score              17415
ELNET_score            15705
GENIE3_score           17416
LLR_score              17358
MI_score               17416
NARROMI_score          14887
PCOR_score             17416
PEARSON_score          17416
TOMSimilarity_score    17416
PLSNET_score           17416
SPEARMAN_score         17416
TIGRESS_score           5869
irp_score              17416
dtype: int64

del result["ARACNE_score"]

X = result.iloc[:,3:]
y = result['Link']
imputer = KNNImputer(n_neighbors=7, weights="uniform")
X_imputed=imputer.fit_transform(X)

X_imputed_abs = pd.DataFrame(X_imputed).abs()
X_imputed_abs.columns = X.columns
print(X_imputed_abs.count())

Type                   17416
Score                  17416
CLR_score              17416
ELNET_score            17416
GENIE3_score           17416
LLR_score              17416
MI_score               17416
NARROMI_score          17416
PCOR_score             17416
PEARSON_score          17416
TOMSimilarity_score    17416
PLSNET_score           17416
SPEARMAN_score         17416
TIGRESS_score          17416
irp_score              17416
dtype: int64

MinMax scaler

scaler = MinMaxScaler()
X_ia_scaled = scaler.fit(X_imputed_abs).transform(X_imputed_abs)
X_ia_scaled = pd.DataFrame(X_ia_scaled)
X_ia_scaled.columns = X_imputed_abs.columns
print(X_ia_scaled)

       Type     Score  CLR_score  ELNET_score  GENIE3_score  LLR_score  \
0       0.0  0.891381   0.182408     0.634058      0.007034   0.902655   
1       1.0  0.093270   0.153364     0.103261      0.012303   0.853097   
2       0.0  0.511216   0.104923     0.085145      0.011892   0.380531   
3       1.0  0.433294   0.257742     0.855072      0.029843   0.886726   
4       0.0  0.889020   0.205640     0.806159      0.056551   0.477876   
...     ...       ...        ...          ...           ...        ...   
17411   1.0  0.682409   0.226718     0.003623      0.013661   0.256637   
17412   1.0  0.465171   0.179210     0.001812      0.032109   0.867257   
17413   1.0  0.192444   0.186172     0.014493      0.003615   0.304425   
17414   1.0  0.463991   0.181123     0.491460      0.013413   0.454867   
17415   1.0  0.675325   0.158612     0.236542      0.014327   0.610619   

       MI_score  NARROMI_score  PCOR_score  PEARSON_score  \
0      0.119899       0.094685    0.019846       0.615117   
1      0.103775       0.056238    0.027622       0.495609   
2      0.066408       0.058254    0.027447       0.503225   
3      0.159029       0.137816    0.034039       0.660524   
4      0.132674       0.104607    0.019225       0.639301   
...         ...            ...         ...            ...   
17411  0.243232       0.128401    0.005651       0.689662   
17412  0.222692       0.272567    0.035931       0.691341   
17413  0.249021       0.113907    0.010416       0.660161   
17414  0.196021       0.088992    0.019909       0.600230   
17415  0.196777       0.116365    0.012516       0.665408   

       TOMSimilarity_score  PLSNET_score  SPEARMAN_score  TIGRESS_score  \
0                 0.106881      0.060727        0.558050       0.094355   
1                 0.051393      0.116439        0.219846       0.095419   
2                 0.054981      0.075575        0.209596       0.181359   
3                 0.080125      0.078906        0.614252       0.001000   
4                 0.064891      0.109174        0.505125       0.000300   
...                    ...           ...             ...            ...   
17411             0.229059      0.068859        0.724409       0.143686   
17412             0.222140      0.079788        0.675824       0.063696   
17413             0.152886      0.042965        0.630406       0.206282   
17414             0.186188      0.076919        0.586525       0.065482   
17415             0.203595      0.087890        0.672217       0.139228   

       irp_score  
0       0.116316  
1       0.077279  
2       0.030559  
3       0.236688  
4       0.178368  
...          ...  
17411   0.108194  
17412   0.177136  
17413   0.069194  
17414   0.023514  
17415   0.032998  

[17416 rows x 15 columns]

PCA

X = X_ia_scaled
tn = ['binding','catalysis' ,'reaction', 'regulation']
y = result['Link'].reset_index(drop=True)

Cumulative explained variance

pca = PCA().fit(X)
plt.plot(np.cumsum(pca.explained_variance_ratio_*100))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

Text(0, 0.5, 'cumulative explained variance')

Plot PCAs

pca = PCA(n_components=4, random_state=42)
X_r = pca.fit_transform(X)
plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset (PC1 vs PC4)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 3], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC4: "+str(round(pca.explained_variance_ratio_[3]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Feature importances

i = 1
for com in pca.components_:
    plt.bar(X.columns,com)
    plt.title("PCA " + str(i)) 
    plt.xticks(rotation = 45, ha="right")
    plt.show()
    i+=1

TSNE

tsne= TSNE(n_components=3,init='pca',
                         random_state=42,
                         n_iter = 500 )
X_r = tsne.fit_transform(X)

plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC2")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC2")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Export

pd.concat([y, X_ia_scaled], axis=1).to_csv("results/"+org+".processed_data.tsv",
                               sep='\t',
                               index=False)

Saccharomyces cerevisiae analysis

org = "scere"
# Reading the preprocess link data.
dfTypeR = pd.read_csv(org+".protein.actions.v11.0.txt.gz",
                      sep="\t",
                      compression="gzip",
                      keep_default_na=True)
# Reading the Seidr file converted in tsv.
dfSf    = pd.read_csv(org+"-network.txt",
                      sep='[;\t]',
                      engine='python',
                      keep_default_na=False,
                      na_values='nan')

# Automatic selection of the colons from Seidr.
selected_col = []
for c in dfSf.columns:
    if "Source" in c or "Target" in c  or "Type" in c or "score" in c :
        selected_col.append(c)
dfSf = dfSf[selected_col]

print("-----------string-action file-----------")
dfTypeR.head()

-----------string-action file-----------

	item_id_a	item_id_b	mode	action	is_directional	a_is_acting	score
0	4932.Q0045	4932.Q0050	binding	NaN	f	f	900
1	4932.Q0045	4932.Q0060	binding	NaN	f	f	540
2	4932.Q0045	4932.Q0065	binding	NaN	f	f	540
3	4932.Q0045	4932.Q0070	binding	NaN	f	f	540
4	4932.Q0045	4932.Q0105	binding	NaN	f	f	804

print("-----------seidr file-----------")
dfSf.head()

-----------seidr file-----------

	Source	Target	Type	ARACNE_score	CLR_score	ELNET_score	GENIE3_score	LLR_score	MI_score	NARROMI_score	PCOR_score	PEARSON_score	TOMSimilarity_score	PLSNET_score	SPEARMAN_score	TIGRESS_score	irp_score
0	Q0010	Q0017	Directed	NaN	5.42318	0.505	1.368760	NaN	0.566362	0.588701	0.006476	0.663444	0.054633	0.024112	0.835139	0.97990	0.440090
1	Q0010	Q0032	Directed	NaN	4.45347	0.170	0.450503	NaN	0.514074	NaN	0.004267	0.507050	0.059810	0.023509	0.797037	0.00050	0.273614
2	Q0017	Q0032	Directed	NaN	6.08534	0.519	0.376180	NaN	0.667295	0.255137	0.002028	0.826293	0.096329	0.016372	0.894470	0.00655	0.412529
3	Q0050	Q0045	Directed	NaN	7.70583	0.013	23.511100	0.436	0.670055	0.154917	0.085100	0.543264	0.084719	0.023748	0.330754	NaN	0.411890
4	Q0055	Q0017	Directed	NaN	1.00634	0.315	6.658460	NaN	0.313977	0.077280	0.014122	0.520888	0.054927	0.018800	0.317671	0.00370	0.255151

dfTypeR_pre = preprocessing(dfTypeR)
link_count = dfTypeR_pre['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

preprocessing
exporting
Done.

to_despreciate = ("ptmod","expression")
dfTypeR_pre = dfTypeR_pre[~dfTypeR_pre['Link'].isin(to_despreciate)]
dfTypeR_pre['Link'] = dfTypeR_pre['Link'].replace(['activation',
                                                   "inhibition",
                                                   "expression-inhibition"],
                                                   'regulation').replace(['binding-activation',
                                                                          'binding-inhibition'],
                                                                          'binding')

# Combine the pandas obj by the columns Source, Target and Type.
result = pd.merge( dfTypeR_pre,dfSf,
                  on=["Source","Target","Type"],
                  how="inner")
# Remove the row duplicates.
result['Type'] = result["Type"].map({"Directed":1,"Undirected":0})
result = result.drop_duplicates()
link_count = result['Link'].value_counts()
indexNamesArr = link_count.index.values
fig, ax = plt.subplots()
plt.bar(indexNamesArr, link_count)
plt.xticks(rotation = 45, ha="right")
plt.show()

binaryTables = {}
for x in result['Link'].value_counts().index.values:
    print(x)
    binaryTables[x] = transformBinaryLinkTable(x,result)

reaction
catalysis
binding
regulation

interaction= result['Source'].str.cat(result['Target'],sep="-")
ij=pd.DataFrame([interaction, result["Link"]]).T.drop_duplicates()
res = ij.groupby('Link')["Source"].apply(' '.join).reset_index()
dict1 = dict()
for row in res.iterrows():
    dict1[res.Link[row[0]]] = set(res.Source[row[0]].split())
venn(dict1)

<AxesSubplot:>

NAs evaluation

print(result.count())

Source                 13877
Target                 13877
Link                   13877
Type                   13877
Score                  13877
ARACNE_score             884
CLR_score              13849
ELNET_score            13873
GENIE3_score           13877
LLR_score               4436
MI_score               13877
NARROMI_score           9812
PCOR_score             13877
PEARSON_score          13877
TOMSimilarity_score    13877
PLSNET_score           13877
SPEARMAN_score         13877
TIGRESS_score           7251
irp_score              13877
dtype: int64

del result["ARACNE_score"]
del result["LLR_score"]

X = result.iloc[:,3:]
y = result['Link']
imputer = KNNImputer(n_neighbors=7, weights="uniform")
X_imputed=imputer.fit_transform(X)

X_imputed_abs = pd.DataFrame(X_imputed).abs()
X_imputed_abs.columns = X.columns
print(X_imputed_abs.count())

Type                   13877
Score                  13877
CLR_score              13877
ELNET_score            13877
GENIE3_score           13877
MI_score               13877
NARROMI_score          13877
PCOR_score             13877
PEARSON_score          13877
TOMSimilarity_score    13877
PLSNET_score           13877
SPEARMAN_score         13877
TIGRESS_score          13877
irp_score              13877
dtype: int64

MinMax scaler

scaler = MinMaxScaler()
X_ia_scaled = scaler.fit(X_imputed_abs).transform(X_imputed_abs)
X_ia_scaled = pd.DataFrame(X_ia_scaled)
X_ia_scaled.columns = X_imputed_abs.columns
print(X_ia_scaled)

       Type     Score  CLR_score  ELNET_score  GENIE3_score  MI_score  \
0       0.0  0.459906   0.416457     0.732143      0.081074  0.308769   
1       0.0  0.459906   0.594422     0.969643      0.239803  0.424727   
2       0.0  0.771226   0.466846     0.891071      0.332114  0.346979   
3       0.0  0.214623   0.329407     0.132143      0.209247  0.281201   
4       0.0  0.214623   0.309281     0.723214      0.201753  0.309069   
...     ...       ...        ...          ...           ...       ...   
13872   0.0  0.770047   0.337940     0.039286      0.076254  0.319452   
13873   1.0  0.362028   0.627795     0.907143      0.138871  0.799344   
13874   1.0  0.362028   0.627795     0.907143      0.138871  0.799344   
13875   0.0  0.976415   0.521855     0.878571      0.608693  0.635716   
13876   0.0  0.824292   0.521855     0.878571      0.608693  0.635716   

       NARROMI_score  PCOR_score  PEARSON_score  TOMSimilarity_score  \
0           0.131183    0.063508       0.727370             0.276220   
1           0.149085    0.087583       0.740082             0.278033   
2           0.277071    0.177148       0.796184             0.287948   
3           0.124246    0.099394       0.690582             0.272007   
4           0.139444    0.016840       0.755282             0.282740   
...              ...         ...            ...                  ...   
13872       0.049473    0.033530       0.475729             0.101155   
13873       0.397787    0.019458       0.959956             0.951825   
13874       0.397787    0.019458       0.959956             0.951825   
13875       0.307724    0.069178       0.857454             0.471613   
13876       0.307724    0.069178       0.857454             0.471613   

       PLSNET_score  SPEARMAN_score  TIGRESS_score  irp_score  
0          0.092101        0.538515       0.604523   0.286979  
1          0.078814        0.653985       0.001400   0.556863  
2          0.130295        0.571202       0.001100   0.562543  
3          0.048021        0.484751       0.624510   0.256637  
4          0.042114        0.424729       0.624510   0.224523  
...             ...             ...            ...        ...  
13872      0.080755        0.317524       0.263499   0.201068  
13873      0.023244        0.952144       0.002650   0.577478  
13874      0.023244        0.952144       0.002650   0.577478  
13875      0.044052        0.880921       0.816541   0.658126  
13876      0.044052        0.880921       0.816541   0.658126  

[13877 rows x 14 columns]

PCA

X = X_ia_scaled
tn = ['binding','catalysis' ,'reaction', 'regulation']
y = result['Link'].reset_index(drop=True)

Cumulative explained variance

pca = PCA().fit(X)
plt.plot(np.cumsum(pca.explained_variance_ratio_*100))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

Text(0, 0.5, 'cumulative explained variance')

Plot PCAs

pca = PCA(n_components=4, random_state=42)
X_r = pca.fit_transform(X)
plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset  (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC2: "+str(round(pca.explained_variance_ratio_[1]*100,2)))
plt.ylabel("% explained variance PC3: "+str(round(pca.explained_variance_ratio_[2]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

plt.title('PCA of dataset (PC1 vs PC4)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 3], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("% explained variance PC1: "+str(round(pca.explained_variance_ratio_[0]*100,2)))
plt.ylabel("% explained variance PC4: "+str(round(pca.explained_variance_ratio_[3]*100,2)))
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Feature importances

i = 1
for com in pca.components_:
    plt.bar(X.columns,com)
    plt.title("PCA " + str(i)) 
    plt.xticks(rotation = 45, ha="right")
    plt.show()
    i+=1

TSNE

tsne= TSNE(n_components=3,init='pca',
                         random_state=42,
                         n_iter = 500 )
X_r = tsne.fit_transform(X)

plt.figure()
colors = sns.color_palette()
lw = 2
plt.title('PCA of dataset (PC1 vs PC2)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 1], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC2")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC1 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 0], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC1")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()
plt.title('PCA of dataset (PC2 vs PC3)')
for color, i in zip(colors, tn):
    plt.scatter(X_r[y == i, 1], X_r[y == i, 2], color=color, alpha=.1, lw=lw,
                label=i[:4]+": "+str(result["Link"].value_counts()[i]))
plt.xlabel("PC2")
plt.ylabel("PC3")
leg = plt.legend(loc="best", shadow=False, scatterpoints=1)
for lh in leg.legendHandles: 
    lh.set_alpha(0.8)
plt.show()

Export

pd.concat([y, X_ia_scaled], axis=1).to_csv("results/"+org+".processed_data.tsv",
                               sep='\t',
                               index=False)